From: Joaquín Casillas Meléndez (jcasilla++at++isoco.com)
Date: 02/12/2003 01:07:17
Tom Flynn wrote:
> On Tue, 11 Feb 2003, [ISO-8859-1] Joaqu?n Casillas Mel?ndez wrote:
>
>
>>I've run perfly for several seconds, and the memory leak is there,
>>consuming MBs at the same rate as with our app. So the problem doesn't
>>seem to be in the app.
>
>
> Hi there...
>
> I just ran two tests. First with perfly and no arguments. I let it run
> for an hour and a half. I checked memory usage using top and an internal
> port of gmemusage. Memory usage didn't change. My second test was with
> perfly town.perfly. It has been running for two hours now, no change in
> memory usage. This was with Performer 3.0 and the NVidia 4191 driver on a
> GeForce3.
>
> Are there steps I should be taking other than what I've listed above in
> order to reproduce the problem you're seeing? Also, what are you using to
> check memory usage?
> -tom
>
Hi Tom,
I followed the same steps. I run perfly without arguments, and after
that I run the town demo. I used top to check memory.
I've just executed perfly -W800,600 during 1 minute (more or less), and
these are several values showed by top during that minute:
PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
5993 xxxx 25 0 152M 22M 22852 R 99.1 4.5 0:10 perfly
5993 xxxx 25 0 186M 57M 58180 R 94.8 11.4 1:33 perfly
5993 xxxx 25 0 200M 71M 72316 R 98.2 14.1 2:06 perfly
As you can see, the RSS has grown a lot in only one minute. Perfly
finally crashes with the message:
PF Fatal/Resource: pfMemory::new() Unable to allocate 36
bytes from
arena 0x41514000.
Try using pfSharedArenaSize() or env PFSHAREDSIZE
to increase the arena size
(currently 65536.00 KBytes) and check for adequate setrlimit()
values and available space on swap (or pfTmpDir()).
I gues I could increase the shared arena size, but I think this won't
solve the problem (we would only get more time to run the app).
The last test I made was to execute perfly in a differente linux
configuration, this time in a Mandrake 9.0 distribution with Performer
3.0 demo edition. The graphics card was from ATI, and the drivers where
those included with the distribution (DRI mode). I've left the town
running last night, and this morning it had crahsed with a message
similiar to the above one. This time memory leakage was fewer than with
the other configuration, but it was still there.
Do you think this has anything to do with kernel drivers, glibc version
or some kind of this stuff? We could try with a different linux
distribution, but not with a different graphics card because that
(GrForce4) has been choosen for the final version.
Thank you for any help.
-- ------------------------------------------------------------------------- Joaqumn Casillas Melindez, R+D Engineer at iSOCO Intelligent Software Components, S.A. http://www.isoco.com Addr: Profesor Beltran Baguena, 4, oficina 305, 46009 Valencia Voice: +34 96 346 71 43 Fax: +34 96 346 71 42
This archive was generated by hypermail 2b29 : Wed Feb 12 2003 - 01:09:24 PST